Fuzzy Approaches To Quality Function Deployment For New Product Design

Fuzzy Sets and Systems 160 (2009) 2620 – 2639
www.elsevier.com/locate/fss
Fuzzy approaches to quality function deployment for new

product design
Liang-Hsuan Chena,∗ , Wen-Chang Koa, b
a Department of Industrial and Information Management, National Cheng Kung University, Tainan, Taiwan, ROC
b Department of Information Management, Kun Shan University, Tainan County, Taiwan, ROC
Received 30 July 2007; received in revised form 29 November 2008; accepted 9 December 2008
Available online 24 December 2008
Abstract
For new product development, quality function deployment (QFD) is a useful approach to maximize customer satisfaction. The
determination of the fulfillment levels of design requirements (DRs) and parts characteristics (PCs) in phases 1 and 2 is an important
issue during QFD processes for new product design. Unlike the existing literature, which mainly focuses on the DRs, this paper
proposes fuzzy nonlinear programming models based on Kano’s concept to determine the fulfillment levels of PCs with the aim of
achieving the determined contribution levels of DRs in phase 1 for customer satisfaction. In addition, to deal with the design risk,
this study incorporates failure modes and effects analysis (FMEA) into QFD processes, and treats it as the constraint factor in the
models. To cope with the vague nature of product development processes, fuzzy approaches are used for both FMEA and QFD. The
applicability of the proposed models in practice is demonstrated with a numerical example.
© 2008 Elsevier B.V. All rights reserved.
Keywords: Design; Fuzzy mathematical programming; Quality function deployment (QFD); Kano’s concept; Failure modes and effects analysis
(FMEA)
1. Introduction
Quality function deployment (QFD) is a useful customer-driven product development tool for translating the needs
of the customer into efficient communication through the various stages of product planning, design, engineering, and
manufacturing to achieve higher customer satisfaction. QFD was conceived in Japan in the late 1960s by Akao and
then disseminated world-wide [1]. It has been successfully introduced in many industries to improve product design,
decision making process, and customer satisfaction [12,16,22]. A typical QFD process consists of four phases. In the
new product design stage, a QFD team is organized to implement the first two phases, i.e., product planning and part
deployment, of QFD processes, which mirror the product design process to achieve great customer satisfaction. The
two phases are closely related at the design stage, since the outcome from the latter phase should make the decisions
from the former phase applicable. However, most of the existing literature only focuses on the first phase of QFD, so
this study will consider the two phases in QFD activities for a new product design.
∗ Corresponding author. Fax: +886 6 236 2162.

E-mail address: lhchen@mail.ncku.edu.tw (L.-H. Chen).
0165-0114/$ - see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.fss.2008.12.003
L.-H. Chen, W.-C. Ko / Fuzzy Sets and Systems 160 (2009) 2620 – 2639 2621
The house of quality (HOQ) is an important tool for QFD activities, containing information on “what”, “how”,
relationship between “what” and “how”, and, the relationship between the “how” factors themselves. For example, for
the development of phase 1, the QFD team determines a set of “what”, i.e., customer requirements (CRs), and “how”,
i.e., design requirements (DRs), that affect CRs, and then constructs the relation strength between CRs and DRs and the
relationships between the DRs themselves. The objective of the first phase is to determine the fulfillment level of DRs
to maximize the customer satisfaction. In phase 2, the DRs are considered as “what”, and parts characteristics (PCs)
are “how” factors designed to maximally achieve the requirement level of DRs. In traditional QFD, the measurements
of importance of “what” and the associated relationships are assessed on a scale system such as 1–3–9, or 1–5–9,
indicating “weak”, “moderate”, and “strong”, respectively [11,16]. Nevertheless, gathering sufficient knowledge and
information about the measurements is usually difficult [9,10]. This has led to the introduction of fuzzy approaches
into QFD studies [5,6,9,10,18]. However, these existing studies still suffer from a number of drawbacks in approaches
and methodologies, as mentioned by Chen and Weng [9].
In practice, effective design processes can lead to the achievement of maximum customer satisfaction. For instance,
in phase 1 of QFD, DRs are able to achieve better effectiveness of the design planning when their performance is highly
correlated with CRs to meet external customer satisfaction. The effect of PCs on DRs in phase 2 is comparable to that
of DRs on CRs. However, the fulfillment levels of DRs and PCs are not necessarily proportionally consistent with CRs’
and DRs’ satisfaction, respectively. Kano et al. [17] proposed that quality performance of a product or service has three
different relationships (i.e., attractive, one-dimensional, and must-be) with customer satisfaction. Some researchers
incorporated Kano’s idea into QFD for assigning weights to different “what” to explain how they impact customer
satisfaction in different ways [15,25,30–32]. Instead, recently Chen and Ko [8] adopted Kano’s concept to categorize
DRs in their fuzzy models. Referring to [8], this study applies Kano’s concept to classify various “how” into three
categories based on their importance ranking, since the performance of various “how” achieves different levels of
external/internal customer satisfaction.
Furthermore, in order to decrease risks inherent in of new product design, the risk analysis of DRs is necessary at the
design stage of new product development. The outcomes of phase 1, i.e., the fulfillment levels of DRs, are applied to
phase 2 as the constraint factors in determining the achievement levels of PCs. With this consideration, failure mode and
effect analysis (FMEA) is applied for risk analysis. The influence of DRs’ risk levels on PCs in phase 2 is considered in
the decision models. FMEA is a systematic technique for identifying, prioritizing and acting on potential failure modes
before the failures occur. Several researchers have discussed FMEA and its applications to QFD [2,13,29,30]. However,
these studies are only limited to descriptive analyses to obtain the quality and resource benefits. The methods to carry
out the aggregation of the QFD and FMEA are not mentioned, and the uncertainty at the new product design stage is
not considered, either. In this study, we construct the fuzzy nonlinear mathematical programming models, referring to
Kano’s concept in phase 1; furthermore, we introduce FMEA into the fuzzy QFD approach, and propose an idea to
link the function relationship between phases 1 and 2 of the QFD in determining the achievement levels of “how” in
phase 2 to maximize the internal and external customer satisfaction under the relevant constraints. An example of a
semiconductor packing case is given to illustrate the proposed models.
In the following section, a fuzzy approach to QFD is introduced. In Section 3, a fuzzy nonlinear model in phase 1
of the QFD process is proposed based on Kano’s concept to categorize “how” into three characteristics. In the fourth
section, the fuzzy FMEA approach is introduced and a fuzzy nonlinear programming model is developed to determine
the achievement degree of PCs, constrained by the need of DRs in phase 1 and the risk rating of PCs according to
the development of FMEA on DRs. An example of a semiconductor packing case is presented to demonstrate our
approaches in Section 5. Finally, the concluding remarks are provided in Section 6.
2. Fuzzy QFD
The first step in the implementation of QFD processes is usually to construct a relation matrix, called a HOQ, for
each phase to define the relationships between “what” and “how”. Based on the information contained in the HOQ,
the achievement priority or level of output variables can be determined. At the design level, the major work in phase 1
of the QFD processes is to determine the achievement priority or level of “how”, i.e., DRs, referring to the importance
of each “what”, i.e., CR, the relationships between CRs and DRs and the relationships between the DRs themselves.
Following the results from the first stage of the QFD process, similar work is performed with DRs and PCs in phase 2.
2622 L.-H. Chen, W.-C. Ko / Fuzzy Sets and Systems 160 (2009) 2620 – 2639
r1,jJ r2,kK
DR1 . . . DRj . . . DRJ PC1 . . . PCk . . . PCK
CR1 k1,1 DR1 k2,1

. . . .
. . . .
. . . .
CRi k1,i R1,ij DRj k2, j R2,jk
. . . .
. . . .
. . . .
CRI k1,I DRJ k2,J
W1,1 . . . W1, j . . . W
1,J W2,1 . . . W2,k . . . W
1,K
Fig. 1. The relationship matrices for phases 1 and 2 of the QFD processes.
Fig. 1 demonstrates the relation matrices of the two phases for the new product design level. Obviously, from the figure
the fulfillment levels of PCs should mean the DRs are appropriate to meet the customer satisfaction. In the figure, R1,i j
denotes the relation level in terms of score between CRi and DR j , and r1, j J is the correlation score between DR j and
DR J inthe first phase of QFD. The notations k and W represent the importance score and rating in the two matrices,
where k = 1.
Considering the relationships between CRs and DRs, Wasserman [33] proposed the normalized relationship value
between CRi and DR j as
J
=1R1,i · r1, j
R1,i j = J J , i = 1, . . . , I ; j = 1, . . . , J. (1)
j=1 =1 R1,i · r1, j
Due to the imprecise nature of the relationships in the stage of new product design planning, Chen and Weng [9]
adopted fuzzy approaches to reformulate the above formulation by describing the relationships, i.e., R1,i and r1, j ,
in terms of linguistic terms, which are characterized by membership functions of fuzzy numbers, i.e., R̃1,i and r̃1, j ,
respectively. Since a fuzzy number can fully and uniquely be represented by its -cuts [20], the calculations in the fuzzy
version can efficiently be performed in terms of -cuts, which are represented as closed intervals of real numbers. For
example, the -cut of the fuzzy set R̃1,i at the level, ∈ [0, 1], can be represented by its lower and upper bounds as
[(R1,i )L , (R1,i )U
], which is defined by
(R1,i )L = inf {x| R̃1,i (x) } and (2a)

x∈[0,1]
= sup {x| R̃1,i (x) },

(R1,i )U (2b)
x∈[0,1]
where R̃1,i (x) is the membership degree of x belonging to R̃1,i . Similarly, the -cuts of r̃1, j also can be obtained.
The -cuts of R̃1,i and r̃1, j are defined as the positive closed intervals in this paper. The -cuts of the fuzzy normalized
relationship can be determined by each -cut of R̃u,i and r̃u, j , u = 1, 2 [20,37,38].
Based on the above concept of numerical intervals, Chen and Weng [9] presented a modified formulation to obtain a
more exact representation of the fuzzy normalized relationship, in which the lower and upper bounds of the membership
function at each -cut are formulated as

J L L
=1 (Ru,i ) (ru, j )
j ) = J ,
L
m(Ru,i J J (3a)
=1 (Ru,i ) (ru,m ) +
U U (Ru,i )L (ru, j )L
m=1 =1
m j
J U U
=1 (Ru,i ) (ru, j )
m(Ru,i U
j ) = J J J
. (3b)
=1 (Ru,i ) (ru,m ) + (Ru,i )U
L L U
m=1 =1 (ru, j )
m j
The above lower and upper bounds of -cuts of modified fuzzy normalized relationship, as u = 1, 2, are aggregated
with the importance of each “what” to determine the fuzzy technical importance ratings of each “how” in the form of
-cuts in phases 1 and 2, respectively.
To deal with the vague nature of certain factors at the design stage, several researchers [5,18,21] have used fuzzy
numbers to express the importance score of CRs in phase 1 of the QFD process. In determining the fuzzy importance
score of CRs, the consensus of customers is important, since the DRs should satisfy the CRs maximally. For doing this,
Karsak [18] used the fuzzy Delphi method to decide the fuzzy importance weight, and we also adopt it to determine
the fuzzy importance score of CRs as a triangular fuzzy number. Appendix A describes the procedures of the fuzzy
Delphi method. Let k̃1,i , a fuzzy subset in [0, 1], represent the final average of the fuzzy importance score of CRi by
experts’ evaluation in the fuzzy Delphi method. For simplification, the fuzzy importance score can be defuzzified so
as to aggregate with the modified fuzzy normalized relationship in determining the fuzzy technical importance ratings
W̃u, j , u = 1, 2, for DR j and PC j , respectively. In this study, we adopt the fuzzy mean (FM) method [23] to carry out
the defuzzification of k̃u,i , u = 1, 2, due to its simplicity and efficiency. The method is expressed as
Q
q=1 q x q
d = Q , (4)
q=1 q
where q and xq denote the membership degree and the representative (pre-calculated) numerical value of the qth
output, respectively. In this study, d is the defuzzification value of interval values [xqL , xqU ], i.e., the qth -cut of fuzzy
number. Following Mabuchi’s idea [24], we define the representative value of the qth output as the average of lower
and upper bounds of the interval by xq = 21 (xqL + xqU ).
For the defuzzification of the fuzzy importance score, the average value of the -cut of k̃u,i at the q level, i.e.,
(ku,i )q = 21 [(ku,i )Lq + (ku,i )U
q ], u = 1, 2, is used as the representative value. Based on the FM method, the k̃u,i can
be defuzzified as

q q · (ku,i )q
ku,i = , u = 1, 2, i = 1, . . . , I. (5)
q q
Furthermore, the fuzzy technical importance ratings W̃u, j , u = 1, 2, for DR j and PC j , respectively, can be
simplified as
I
I

i=1 ku,i · m(Ru,i j ) i=1 ku,i · m(Ru,i j )
L U
(W̃u, j ) = [(Wu, j ) , (Wu, j ) ] =
L U
I , I . (6)

i=1 ku,i i=1 ku,i
The resulting W̃u, j is then employed to find the optimal fulfillment level of each DR and PC in phases 1 and 2 of the
QFD process.
3. Fuzzy formulations of phase 1
Using the fuzzy technical importance ratings of DRs, Chen and Weng [9] proposed a fuzzy linear model, in which
the level of fulfillment percentage of the DR j , x1, j (∈ [0, 1]), is determined. If x1, j = 0, it indicates that only a basic
DR is needed for DR j , so that no more effort and cost are needed. In Chen and Weng’s model, the fulfillment levels of
DRs are determined to maximize the total customer satisfaction. From this perspective, the fulfillment levels of DRs are
Very satisfied
Satisfaction One-Dimensional
Attractive Performance
Not at all Fully
Must-be
Very dissatisfied
Fig. 2. Kano’s conceptual model.
related to product performance, and DRs with the higher technical importance rating will have a greater contribution to
total customer satisfaction. In other words, even DRs which have equal fulfillment levels may achieve different levels
of customer satisfaction. To classify DRs, this paper employs Kano’s concept to categorize DRs into different classes
to reinforce the DRs with higher technical importance ratings. Kano et al. [17] proposed a useful method to quantify
the relationship between the quality performance of a product or service and customer satisfaction. Three kinds of
conceptual relationships are demonstrated in Fig. 2, including attractive, one-dimensional, and must-be relationships.
They are described as follows. The DRs with an attractive relationship have the property that increasing performance
of the product in terms of the fulfillment levels can produce more customer satisfaction than with other DRs. The
one-dimensional relationship is a linear function between performance and satisfaction. The relationship is must-be if
an essential performance of the product is sufficient to satisfy the customers’ need.
As mentioned above, the fulfillment level of each DR is related to product performance, and its fuzzy technical
importance rating W̃1, j is closely associated with customer satisfaction. It is logical to categorize DRs into three
Kano’s classes based on the fuzzy technical importance ratings. The membership function of the fuzzy technical
importance rating of each DR is determined by (6). In short, the categorization of DRs can be achieved based on the
rank of the fuzzy technical importance ratings, W̃1, j , of DRs. Based on (3) and (6), W̃1, j is convex and normal, and
therefore is a fuzzy number. The same consideration is also applied to PCs in phase 2. For performing the ranking of
W̃u, j , the defuzzification method, i.e., FM, is applied for each W̃u, j , u = 1, 2.

q q · 21 [(Wu, j )Lq + (Wu, j )U
q ]
Wu, j = . (7)
q q
After ranking the fuzzy technical importance ratings in phase 1, all DRs can be categorized into three classes in
accordance with Kano’s concept. Let 1-class DRs be ones with the higher defuzzified value with respect to the reference,
and 3-class be the reverse. Obviously, the 1-class DRs can increase customer satisfaction more than the DRs in the
other classes. In other words, increasing the performance level by 1-class DRs, and reducing that by 3-class DRs due
to budgetary limitations, can enhance the total customer satisfaction. Based on this consideration, we treat the 1-, 2-,
and 3-class DRs as being attractive requirements, one-dimensional, and must-be, respectively.
In order to incorporate the above concepts into the fuzzy model, we modify the decision variable x1, j representing
q
the performance level of DR j as x1,j j , where q j = 0.5, 1, and 2, when the DR has the attractive, one-dimensional, and
must-be feature, respectively, due to x1, j ∈ [0, 1]. In addition, the increased unit cost to achieve the fulfillment level is
represented as a fuzzy number, C̃1, j , to reflect its fuzzy nature at the design stage. With the fulfillment percentage of
DR j , a corresponding percentage of the increment unit cost is required to enhance the quality of the product or service.
However, the total increased unit cost cannot exceed a cost constraint. Besides the cost limitation, the impact on customer
satisfaction of various DRs is prioritized, and business competition and technological difficulties are also considered in
q q
the model. If DRs is preferred to DR p in terms of customer satisfaction, a constraint of W̃1,s · x1,ss − W̃1, p · x1,pp 0 is
needed. The concept of -cuts is adopted to solve the fuzzy nonlinear programming model in phase 1 of the QFD process.
The lower and upper bounds of -cuts of W̃1, j and C̃1, j are placed in the model to find the lower and upper values
q
of x j j and the total customer satisfaction at each level, respectively. Consequently, a fuzzy nonlinear programming
model to maximize customer satisfaction can be formulated as

J
q
(Z 1 )L = max (W1, j )L · x1,j j
j=1

J
q
· x 1, j B1 ,
(C1, j )U j
s.t.
j=1
q q
(W1,s )L · x1,ss − (W1, p )U
· x 1, p 0,
p
q
0 1, j x1,j j 1, j 1, ∀ j,
s, p ∈ {1, 2, . . . , J }, (8a)

J
q
(Z 1 ) = max
U
· x 1, j
(W1, j )U j
j=1

J
q
s.t. (C1, j )L · x1,j j B1 ,
j=1
q q
· x 1,s − (W1, p ) · x 1, p 0,
(W1,s )U s L p
q
0 1, j x1,j j 1, j 1, ∀ j,
s, p ∈ {1, 2, . . . , J }, (8b)
where B1 is the budget limitation, and (Z 1 )L as well as (Z 1 )U

represent the lower and upper bounds of objective values,
i.e., total customer satisfaction in phase 1, at each level, respectively. In addition, 1, j and 1, j denote the minimum
required level due to the business competition and the maximum level due to technical difficulty, respectively, and
delimit the possible range of the fulfillment level of one DR. The ways to determine 1, j and 1, j could be different
for various business units. In general, the two values can be provided by technical marketing and engineering experts
who involve in the QFD team for the new product design using benchmarking or the group decision-making approach,
such as the fuzzy Delphi method [18,19].
When the application ends, the possible performance level of one DR with the attractive feature can be increased,
while that with the must-be feature can be reduced for the increment of the total customer satisfaction under the
budgetary limitation. After solving the above model at various ∈ [0, 1], the lower and upper bounds of the objective
q
function and each decision variable x j j can be obtained, and then the associated membership functions are determined.
4. Fuzzy formulations of phase 2
To reduce design risk, designers usually perform risk analysis of DRs, and then the risk is incorporated into the
design of PCs in phase 2 during the design stage of new product development. FMEA is an effective measure to provide
information for making risk management decisions. This study also adopts this approach to undertake risk analysis
of each DR in a fuzzy environment. A conventional form of FMEA includes (i) the design function of parts, (ii) the
potential failure mode (categories of failure), (iii) the potential effects of failure (measured by the severity index), (iv)
Table 1
Scales for severity.
Severity Rating
Remote 1
Low 2
3
Moderate 4
5
6
High 7
8
Very high 9
10
Table 2
Scales for occurrence.
Frequency of occurrence Rating Possible failure rate
Remote 1 < 1 : 20, 000

Low 2 1:20,000
3 1:10,000
Moderate 4 1: 2,000
5 1:1,000
6 1:200
High 7 1:100
8 1:20
Very high 9 1:10
10 1:2
Table 3
Scales for detectability.
Detectability Rating Probability of detectability (%)
Remote 1 86–100
Low 2 76–85
3 66–75
Moderate 4 56–65
5 46–55
6 36–45
High 7 26–35
8 16–25
Very high 9 6–15
10 0–5
the potential causes of failure (measured by the occurrence (frequency) index), (v) the detection method (measured by
the detectability index), and (vi) the risk priority number (RPN). The RPN is used to evaluate the risk level of a part’s
failure mode at the design stage, and is determined by the multiplication of three characteristic failure mode indices,
i.e., the severity of the potential failure (S), the frequency of potential failure (O), and the detectability index (D) as
RPN = S × O × D, in which the three indices are rated on the same scale level, such as a 10-point system, to identify
the various levels of risk. Tables 1, 2 and 3 present the scales for the S, O and D indexes [27].
In general, more than one failure mode could happen for each design function (or DR in QFD) of a part in FMEA,
and each failure mode may be described by more than one effect of failure. In addition, more than one cause of failure,
which could be detected by several detection methods, could result in one effect of failure. The detectable degree of
each cause of failure is measured by the detectability index (D). Each failure mode is evaluated by RPN in terms of
the three indices, i.e., S, O, and D. Usually, those indices are assessed subjectively and qualitatively in terms of natural
language; therefore, sometimes conventional FMEA cannot easily rate the failure indices. As an example, the frequency
of potential failure could be assessed to be close to 1:50, for which the existing measurement system cannot give a
suitable rating. Accordingly, a linguistic term, such as “medium high”, can be adopted to describe such a measurement.
Some researchers have proposed fuzzy approaches to deal with the above problems [14,27,28,36]. All of these studies
used adequate experts’ knowledge to build up the fuzzy rule base for reasoning to obtain an acceptable RPN. However,
integrating the related experiences is not easy during the new product design stage, so that the development of a suitable
inference system could be difficult. In this study, we make the risk analysis of DRs by performing FMEA at the design
stage. To describe the assessments meaningfully, the three failure indices are evaluated in linguistic terms, represented
as fuzzy sets, such as “R (remote)”, “L (low)”, “M (moderate)”, “H (high)”, and “VH (very high)”. The fuzzified RPN
of each DR can be expressed as follows:
(
RPN) j = max( S̃r ⊗ Õs ⊗ D̃t ) j , j = 1, 2, . . . , J, (9)
r,s,t
where S̃r , Õs , and D̃t are defined as the fuzzy subsets of [0,1], and the symbol ⊗ represents the fuzzy multiplication
operator. The ( RPN) j is determined as the maximum of the three indices’ product from all sets of S, O, and D, if there
is more than one set, to represent the fuzzy risk priority level of the jth DR. Furthermore, considering the different
importance of each failure index, various indices can have different weights to determine ( RPN) j . This study uses a
fuzzy ordered weighted geometric averaging (FOWGA) operator [35] to incorporate different weights of failure index
in (9).
mAn FOWGA operator is used to aggregate m(> 1) fuzzy sets, and the formulation is expressed as f (ã1 , ã2 , . . . , ãm ) =
( b̃ ) vi , where b̃ is the ith largest one in m fuzzy numbers ã , k = 1, . . . , m; v is the weight of the b̃ and
i=1
m
i i k i i
i=1 vi = 1, vi ∈ [0, 1]. Following FOWGA, (9) can be reformulated as

3
(
RPN) j = f ( S̃, Õ, D̃) j = max (b̃i,( j) )vi , (10)
r,s,t
i=1
where b̃i,( j) is the ith largest one among the set ( S̃r , Õs , D̃t ). In (10), the weighting vector of v = (v1 , v2 , v3 )T can be
determined according to designers’ or QFD team members’ experience. In order to determine the membership function
of (
RPN) j from (10), we present the solution procedure as follows. For generalization, assume there is more than one
set of ( S̃r , Õs , D̃t ) for the jth DR.
(1) Define the linguistic terms to evaluate the three failure indices, such as “R (remote)”, “L (low)”, “M (moderate)”,
“H (high)”, and “VH (very high)”. These terms are represented as fuzzy sets, and have the relations among them
as R ≺ L ≺ M ≺ H ≺ VH, where ≺ denotes “approximately less than”.
(2) Order the three failure indices of each set ( S̃r , Õs , D̃t ) j for the jth DR using the predetermined linguistic terms;
the b̃i,( j) in sequence for (10) is then obtained.
(3) Use the lower and upper values of b̃i,( j) for each -cut as well as the weighting vector to calculate ((bi,( j) )L )vi and
((bi,( j) )U vi
) for each failure index. 3
(4) Calculate the lower and upper values of ( RPN) j for each -cut from each set of ( S̃r , Õs , D̃t ) j by i=1 ((bi,( j) )L )vi
3 v
3 v
3 v
and i=1 ((bi,( j) ) ) i , i.e., (RPN j ) = i=1 ((bi,( j) ) ) i and (RPN j ) = i=1 ((bi,( j) ) ) i , respectively.
U L L U U
(5) Apply (RPN j )L and (RPN j )U from each set of ( S̃r , Õs , D̃t ) j to the FM method in order to determine the maximum

(R PN) j . Suppose that the set ( S̃r , Õs , D̃t ) j causes the maximum fuzzy risk priority number (RPN ) j having the
lower and upper values as

3
(RPNj )L = ((bi,( j) )L )vi and (11a)
i=1
r ,s ,t
3
(RPNj )U
=
vi
) .
((bi,( j) )U (11b)
i=1
r ,s ,t
) j is denoted as
The defuzzified value of (RPN

q q (P R N j )q
RPN j = , j = 1, . . . , J, (12)
q q
where (RPNj )q = 21 [(RPNj )Lq + (RPNj )U

q ].
) j represents the risk level of the jth DR, and is applied to phase 2 for parts design
The defuzzified value of (RPN
deployment in determining the fuzzy risk rating of PCs with the consideration of the fuzzy normalized relationship.
The formulation is expressed as
J
J

j=1 RPN j · m(R2, jk ) j=1 RPN j · m(R2, jk )
L U

( Ri k ) = [(Ri k ) , (Ri k ) ] =
L U
J , J . (13)

j=1 RPN j j=1 RPN j
The fuzzy risk rating of each PC is used in phase 2 to determine the fulfillment level of each PC in the design level to
satisfy the DRs and finally maximize customer satisfaction.
To develop the fuzzy mathematical model for phase 2 of the QFD process, the importance scores of the DRs in
the second phase should be determined. In the existing literature, the importance ratings (W1, j ) of the DRs in phase 1
are adopted as the importance scores (k2, j ) for the deployment of phase 2 in QFD processes [26]. In the deployment
processes, the fulfillment levels of PCs in phase 2 should make those of DRs in phase 1 applicable. In this study,
we proposed a weighted-average method to determine the importance score using the importance ratings and the
achievement levels of DRs in phase 1 as follows:
q q
(k2, j ) = 21 [(W1, j )L (x1,j j )L + (W1, j )U j U
(x 1, j ) ], j = 1, . . . , J. (14)
Similarly, for simplifying the importance scores in phase 2, we applied (5) (u = 2) to obtain a defuzzified value. The
-cut of the PCs’ fuzzy importance ratings can be determined by (6) as u = 2. The same as in phase 1, the fuzzy
importance ratings of PCs can be ranked by (7), and then be classified into three Kano’s categories to construct the
fuzzy nonlinear model to maximize the satisfaction of design planning in phase 2.
Extending Model (8) by including the risk analysis of DRs, this study proposed a fuzzy nonlinear programming
qk
model to determine the fulfillment level of the PCs in phase 2. Similar to Model (8), the notation of x2,k is defined as
the fulfillment level of the kth PC in phase 2. By performing FMEA on DRs, the fuzzy risk ratings of PCs are obtained
by (13) and used as the constraint in the proposed model. In addition, the determination of the fulfillment levels of the
PCs has to make DRs applicable to achieve customer satisfaction. The contribution to customer satisfaction by DR j
K qk
due to the design satisfaction of PCs in phase 2 can be expressed as k2, j [ k=1 m( R̃2, jk )x 2,k ], which should be greater
q
than or equal to the contribution of DR j , W̃1, j x̃1,j j , in phase 1 to realize customer satisfaction. Moreover, the budget
limitation, preemptive priority, and technological ability for PCs are also taken into account in the proposed model.
Taking the above considerations as the constraints as well as using the PCs’ fuzzy importance ratings of (6) (u = 2)
qk
and the decision variables, x 2,k , as the objective function, a fuzzy nonlinear programming model is formulated in the
-cut form as

K
q
(Z 2 )L = max (W2,k )L · x2,k
k
k=1
K
q
qk
s.t. k2, j m(R2, L
jk ) · x2,k W1, j x1,j j , j = 1, . . . , J,
k=1

K
q
· x 2,k H,
(Ri k )U k
k=1

K
q
· x 2,k B2 ,
(C2,k )U k
k=1
q q
(W2,s )L · x2,ss − (W2, p )U
· x 2, p 0,
p
q
0 x 2,k
k
2,k 1, ∀k,
s, p ∈ {1, 2, . . . , K }, (15a)

K
q
= max
(Z 2 )U · x 2,k
(W2,k )U k
k=1
K
q
qk
s.t. k2, j m(R2, U
jk ) · x2,k W1, j x1,j j , j = 1, . . . , J,
k=1

K
q
(Ri k )L · x2,k
k
H,
k=1

K
q
(C2,k )L · x2,k
k
B2 ,
k=1
q q
· x 2,s − (W2, p ) · x 2, p 0,
(W2,s )U s L p
q
0 x2,k
k
2,k 1, ∀k,
s, p ∈ {1, 2, . . . , K }. (15b)
In the above models, the objective functions are to find the maximum satisfaction of DR in phase 2. The values of
represent the lower and upper bounds of objective values, respectively, at each -cut. In the constraints,
(Z 2 )L and (Z 2 )U
H is a threshold of risk; B2 is a budgetary limitation; C̃2,k and 2,k are the increased unit cost to achieve the fulfillment
level and technological difficulty of the PCs, respectively. These parameters are determined by the QFD team. In
addition, for purposes of simplification and application, the crisp contribution level of each DR in phase 1 is taken as
q q
the action by the QFD team and is adopted in the model. In Model (15), W1, j x1,j j is a defuzzified value of W̃1, j x̃1,j j
. q
and then calculated by (14) and (5), so that W1, j x1,j j is equal to k2, j
5. An illustrative example
In this section, a semiconductor packing case of the turbo thermal ball grid array (T2 -BGA) package is used to
exemplify the applicability of the proposed models. T2 -BGA is a member of the ball grid array (BGA) family in
microelectronics packaging, and it can provide excellent thermal and electrical performance, since a heat slug is inserted
into the molding compound of a plastic BGA (PBGA). The heat slug has the advantages of high heat dissipation and
good ground shielding to reduce the electromagnetic interference (EMI) and electromagnetic compatibility (EMC)
effects between adjacent traces; however, it also produces some problems due to the more complex structure of the
package when compared to PBGA [7]. Fig. 3 displays the cross-sections of PBGA and T2 -BGA.
Chip Goldwire Heatslug Molding compound
Solder ball Substrate

PBGA T2-BGA
Fig. 3. The cross-sections of PBGA and T2 -BGA.

DR1 DR2 DR3 DR4 DR5
CR1 M M L
CR2 L L
CR3 L M
CR4 H L
CR5 L
Fig. 4. The phase 1 HOQ for a T2 BGA package.
1 VL L M H VH
Membership degree
0
.2 .3 .4 .5 .6 .7 .8 1
Strength scale
Fig. 5. The membership functions of linguistic variables in the HOQ.
5.1. Phase 1 of the QFD process
Based on the above considerations of T2 -BGA, a QFD team collects five CRs and proposes five DRs in phase 1 of
the QFD process. The fuzzy relations, R̃1,i , between CRs and DRs and the relevant relations r̃1, j among DRs are
shown in the HOQ in Fig. 4. In phase 1, the CRs are “package profile” (CR1 ), “thermal performance” (CR2 ), “electrical
performance” (CR3 ), “reliability” (CR4 ), and “co-planarity” (CR5 ), and the five DRs are “heat slug exposed area”
(DR1 ), “material attached to heat slug” (DR2 ), “height of heat slug” (DR3 ), “copper pattern” (DR4 ), and “molding
flow” (DR5 ). The importance levels of CRs and their relationships are denoted as “very low (VL)”, “low (L)”, “medium
(M)”, “high (H)” or “very high (VH)”, which are translated into the triangular fuzzy numbers as a 3-element set, (0,
0, 0.2), (0, 0.2, 0.4), (0.3, 0.5, 0.7), (0.6, 0.8, 1), (0.8, 1, 1), respectively, with membership functions in Fig. 5. As an
example, R̃1,i = (0.3, 0.5, 0.7) is formulated as
⎧
⎪
⎪ (R1,i − 0.3)
⎨ (0.5 − 0.3) , 0.3 R1,i 0.5,
⎪
R̃1,i (R1,i ) =
⎪
⎪ (0.7 − R1,i )
⎪
⎩ , 0.5 R1,i 0.7.
(0.7 − 0.5)
The middle value in the 3-element set representing the membership function of a fuzzy number is the most likely one
with the membership degree equivalent to 1, such as H (0.8) = 1. In this study, the system of 0–0.2–0.5–0.8–1 is used
to express the most likely value of the linguistic terms on VL–L–M–H–VH. Based on this system, the fuzzy Delphi
method is used to obtain the fuzzy importance score k̃1,i , and finally they are calculated as (0.038, 0.238, 0.438), (0.263,
0.463, 0.663), (0.113, 0.313, 0.513), (0.55, 0.75, 0.925), (0, 0.15, 0.35) for the five CRs, respectively. The relevant
data and computational procedures are listed in Appendix B. Actually, the lower and upper bounds of each -cut can
be represented as the functions of (∈ [0, 1]). As an example, the -cut of the membership function (0, 0.15, 0.35) of
k̃1,5 is expressed as [(k1,5 )L , (k1,5 )U
] = [0 + 0.15, 0.35 − 0.2]. For simplification, we can defuzzify k̃1,i by (5) as
0.157, 0.286, 0.2, 0.447, and 0.11 for the five CRs, respectively. Furthermore, considering the linguistic relationships
in Fig. 4, the fuzzy normalized relationship at each -cut can be calculated using (3). Without any biases, the levels
in this paper are evenly distributed in [0, 1]. Let denote the th level, and = /n, ∈ {0, 1, . . . , n}, such that
the distance between each two adjacent levels is equal, i.e., − −1 = 1/n, 1. The lower and upper bounds
of the fuzzy technical importance rating (W̃1, j ) , j = 1, 2, . . . , J , at each level, is obtained by (5) to determine the
importance priority of each DR, as listed in Table 4. They are applied to (7) and (8) to categorize the DRs based on the
three basic characteristics of Kano’s concept.
By applying six -cuts, (W̃1, j ) , in Table 4 to (7), the defuzzification values W1, j , j = 1, . . . , 5, are 0.167, 0.232,
0.056, 0.329, and 0.095, respectively. In order to assign each DR into the associated class based on the measure, a
relative scale is used in the following:
Wu, j − min{Wu, j }
Wu, j = , u = 1, 2; j = 1, 2, . . . , J. (16)
max{Wu, j } − min{Wu, j }
In this case, the resulting scales are 40.6%, 66.4%, 0%, 100%, and 14.1%. For the assignments, we adopt a three-interval
system, i.e., 0–33.3–66.7–100%, in phase 1 based on the assumption that the quantity of the relative difference is closely
related to a DR’s feature in the interval manner. Following this rule in phase 1 of the QFD, the fourth requirement DR4
is considered the one with the attractive feature; DR1 and DR2 have one-dimensional features; DR3 and DR5 have
must-be features.
In addition to incorporating Kano’s concept, the fuzzy nonlinear model also considers the business competition,
technical difficulty, and required cost associated with the complete performance level of each DR, as listed in Table
5. Furthermore, to be in accordance with Kano’s categorization, the corresponding business competition is reset as
1,1 = 1,1 = 0.1, 1,2 = 1,2 = 0.1, 1,3 = 0.01, 1,4 = 0.32, and 1,5 = 0.25, in constructing the fuzzy nonlinear
model. The budgetary total increased cost is limited to 1 unit, i.e., B1 = 1.0 in Model (8). The model assumes that the
QFD team has prioritized DRs as DR4 DR1 DR2 and DR3 DR5 for the design preference, where “” means
“is more preferred than”.
Applying fuzzy technical importance ratings, fuzzy increment cost, and the associated constraints to Model (8) at
several levels, the performance level of each DR and the total customer satisfaction degree (the objective function
value) can be obtained. In phase 1 of this example, the settings of levels less than 0.4 will produce infeasible regions
in solving the fuzzy nonlinear model. The outcomes of the total customer satisfaction degree and fulfillment levels of
each DR for Model (8) based on five levels (0.5, 0.6, 0.7, 0.8, 0.9, 1) are listed in Table 6. The decision variable
x 1, j = 100% denotes complete fulfillment of DR j . In this example, DR4 should have the best quality level in order to
achieve the CRs’ satisfaction degree. On the other hand, DR2 is smaller than the others, due to the high incremental
cost, as shown in Table 5.
For application purposes, the crisp performance level can be taken as the action by the QFD team in phase 1. For
doing this, we can adopt a crisp performance level of each DR at a particular level that the QFD team accepts. In
fact, in fuzzy term, the level can be interpreted as the confidence degree [4,34]. Based on the acceptable confidence
degree, the crisp performance level of DR1, j from the fuzzy nonlinear model can be determined as
q q
j ) = (x 1, j ) + (1 − )(x 1, j ) ,
j L j U
(x1, (17)
q q
are the lower and upper bounds of DR1, j at the level, respectively, and the index is used
where (x1,j j )L and (x1,j j )U
to reflect the QFD team’s attitude to the lower or upper bound. Suppose that the QFD team’s attitude is neutral and
level = 0.9 is accepted for the design purpose. The crisp performance levels of the five DRs at = 0.9, (x1, )
j =0.9 ,
j = 1, . . . , 5, are then 0.985, 0.153, 0.27, 1.0, and 0.16, respectively.
Table 4
The lower and upper bounds of (W̃1, j ) at different levels.
L
W1,1 U
W1,1 L
W1,2 U
W1,2 L
W1,3 U
W1,3 L
W1,4 U
W1,4 L
W1,5 U
W1,5
1 0.167 0.167 0.228 0.228 0.055 0.055 0.328 0.328 0.092 0.092
0.8 0.134 0.200 0.202 0.257 0.047 0.063 0.295 0.361 0.069 0.116
0.6 0.100 0.234 0.178 0.287 0.040 0.071 0.261 0.395 0.049 0.141
0.4 0.067 0.267 0.155 0.319 0.034 0.080 0.228 0.428 0.031 0.168
0.2 0.033 0.300 0.136 0.352 0.028 0.088 0.195 0.462 0.014 0.196
0 0 0.334 0.118 0.386 0.023 0.097 0.161 0.495 0 0.224
Table 5
The cost, business competition, and technical difficulty for each DR.
DR j C̃1, j -cut 1, j 1, j
DR1 (0.2, 0.3, 0.4) [0.2 + 0.1, 0.4 − 0.1] 0.1 1.0
DR2 (0.8, 0.9, 1.) [0.8 + 0.1, 1 − 0.1] 0.1 1.0
DR3 (0.4, 0.5, 0.6) [0.4 + 0.1, 0.6 − 0.1] 0.1 1.0
DR4 (0.3, 0.4, 0.5) [0.3 + 0.1, 0.5 − 0.1] 0.1 1.0
DR5 (0.1, 0.2, 0.3) [0.1 + 0.1, 0.3 − 0.1] 0.4 1.0
Table 6
The outcomes of the total customer satisfaction degree and fulfillment level of each DR.
Z 1L Z 1U L
x1,1 U
x1,1 L
x1,2 U
x1,2 L
x1,3 U
x1,3 L
x1,4 U
x1,4 L
x1,5 U
x1,5
1 0.559 0.559 1 1 0.151 0.151 0.515 0.515 1 1 0.4 0.4

0.9 0.508 0.609 0.970 1 0.1 0.206 0.568 0.466 1 1 0.4 0.4
0.8 0.452 0.658 0.792 1 0.1 0.258 0.619 0.419 1 1 0.4 0.4
0.7 0.396 0.711 0.586 1 0.1 0.1 0.682 0.642 1 1 0.4 0.684
0.6 0.348 0.777 0.384 1 0.1 0.1 0.742 0.650 1 1 0.4 0.788
0.5 0.281 0.853 0.364 1 0.1 0.1 0.816 0.651 0.618 1 0.4 0.909
5.2. Phase 2 of the QFD process
As described above, considering risk reduction in design stage, the potential failure modes and effects for risk analysis
(FMEA) is performed for the DRs for applying to phase 2. An analysis of the T2 BGA package is listed in Table 7, in
which the three failure indices are described in linguistic terms, “R (remote)”, “L (low)”, “M (moderate)”, “H (high)”,
and “VH (very high)”. For the following resolution procedure, these linguistic terms should be translated to fuzzy
numbers. Their definitions are (0, 0, 1, 2), (1, 2.5, 2.5, 4), (3, 5, 5, 7), (6, 7.5, 7.5, 9), and (8, 9, 10, 10), respectively, as
shown in Fig. 6.
To obtain the fuzzy risk priority number (RPN), the upper and lower bounds of -cuts of the three fuzzy failure indices
S̃r , Õs , D̃t , and their weights should be determined beforehand. Suppose that the set of weights is v = {0.7, 0.21, 0.09},
determined by the QFD team. The fuzzy RPN of each DR is constructed by (11), and defuzzified by (12) for risk
ratings in phase 2. The outcomes from (12) can be normalized as 0.189, 0.24, 0.205, 0.189, and 0.177 for the five DRs,
respectively.
From the outcomes of phase 1, the fuzzy technical importance rating and fulfillment level of each DR are applied to
(14) and defuzzified by (5) (u = 2) to determine the importance score in phase 2 of the QFD. After the normalization,
they are calculated as 0.274, 0.057, 0.034, 0.584 and 0.05 for the five DRs, respectively. Moreover, the QFD team
investigates the characteristics of the principal parts or subsystems to deploy in phase 2 of the QFD. There are five PCs,
Table 7
T2 -BGA package FMEA.
DR j Design function Potential failure Potential effects of S Potential causes O Detection method D
of part mode failure of failure
Heat slug exposed area Distortion of EMC overflow H Packing/Transportation M Profile inspection L
the heat slug
Warpage of the EMC overflow H Heat slug forming M Profile inspection M
heat slug
Plating layer Package looks defect H Plating solution M Surface inspection L
peeling off contaminated
Material attached to Molding failure Package profile defect VH Turbulence M Simulation H

heat slug
Height of heat slug Plating layer Package looks defect M Plating solution L Surface inspection R
peeling off contaminated
Plating thickness Substrate damage H Stress of the H. S. M Stress measurement H
inconsistent
EMC overflow H Heat slug height L H. S. Height M
out of spec. inspection
Unusual height Package looks defect M Unsuitable packaging L Eyes inspection R
Short circuit M Burr L Surface inspection R
Copper pattern Popcorn Reliability test failed H Pattern distribution M Reliability test M
Molding flow Popcorn Reliability test failed H Turbulence M Simulation L
Remote Low Moderate High Veryhigh

1
Membership degree
0
0 1 2 3 4 5 6 7 8 9 10
Ranking of linguistic terms
Fig. 6. The membership functions of linguistic terms in the FMEA.
including “profile of heat slug” (PC1 ), “leg shape of heat slug” (PC2 ), “plating quality of heat slug” (PC3 ), “jointability
between attached material and molding compound” (PC4 ), and “jointability among attached material, heat slug and
substrate” (PC5 ), as shown in Fig. 7. Similar with phase 1 of the QFD, considering the relationships between DRs and
PCs as well as among the PCs in Fig. 7 and the membership functions of linguistic relationships in Fig. 5, the fuzzy
normalized relationship at each -cut can be obtained using (3). The lower and upper bounds of the fuzzy importance
rating of PCs, (W̃2,k ) , k = 1, 2, . . . , K , at each level, are obtained by (6) to determine the importance priority of
each PC, as listed in Table 8. Same as phase 1, W̃2,k is applied to (7) and Model (15) to categorize PCs based on three
basic characteristics of Kano’s category for building up fuzzy nonlinear models in phase 2.
By applying six -cuts, (W̃2,k ) , in Table 8 to (7) and then using (16) to normalize them, the resulting scales are
100%, 23.9%, 0%, 4.9%, and 30.3%. For the assignments, if we follow the three-interval system as phase 1, PC1 is
considered the one with the attractive feature; however, the other four PCs will have must-be features. This consequence
seems unsatisfactory. Further examining the relative difference measures, the above assignments seem to be stringent
for PC2 and PC5 , since they are much larger than PC3 and PC4 . Based on such considerations, we classify PC2 and PC5
L M
L L L
PC1 PC2 PC3 PC4 PC5
DR1 H L
DR2 M H
DR3 H M
DR4 M L
DR5 H L
Fig. 7. The phase 2 HOQ for a T2 BGA package.
Table 8
The lower and upper bounds of (W̃2,k ) at different levels.
L
W2,1 U
W2,1 L
W2,2 U
W2,2 L
W2,3 U
W2,3 L
W2,4 U
W2,4 L
W2,5 U
W2,5
1 0.446 0.446 0.117 0.117 0.012 0.012 0.032 0.032 0.138 0.138
0.8 0.380 0.519 0.095 0.140 0.011 0.014 0.026 0.040 0.105 0.177
0.6 0.321 0.600 0.074 0.164 0.009 0.016 0.020 0.049 0.076 0.221
0.4 0.270 0.685 0.055 0.189 0.007 0.018 0.015 0.059 0.052 0.269
0.2 0.226 0.773 0.039 0.212 0.006 0.020 0.011 0.070 0.031 0.322
0 0.187 0.859 0.024 0.235 0.005 0.022 0.008 0.083 0.014 0.375
Table 9
The fuzzy risk ratings of PCs in various .
Ri 1L Ri 1U Ri 2L Ri 2U Ri 3L Ri 3U Ri 4L Ri 4U Ri 5L Ri 5U
1 0.204 0.204 0.254 0.254 0.074 0.074 0.127 0.127 0.147 0.147
0.8 0.176 0.236 0.219 0.292 0.063 0.086 0.103 0.156 0.125 0.170
0.6 0.150 0.270 0.186 0.331 0.053 0.098 0.081 0.189 0.106 0.196
0.4 0.128 0.306 0.156 0.371 0.044 0.110 0.062 0.226 0.088 0.224
0.2 0.108 0.342 0.130 0.412 0.035 0.121 0.046 0.269 0.073 0.253
0 0.091 0.378 0.106 0.451 0.027 0.133 0.032 0.317 0.060 0.283
as ones with one-dimensional features. In addition, the fuzzy risk ratings of each PC at each -cut can be calculated by
(13) and are presented in Table 9.
The resulting fuzzy importance and risk level ratings of PCs with the associated constraints listed in Table 10 are
applied to Model (15) at several levels to solve the fuzzy nonlinear model. The fulfillment level of each PC and the
total DR satisfaction degree (the objective function value) can then be obtained. In phase 2 of the QFD processes, the
budgetary total cost is limited to 1.5 units, i.e., B2 = 1.5 in the model; the risk threshold H is bounded to 1 in this
case. For phase 2, the set of DRs’ importance scores by (14) and (5) are {0.141, 0.029, 0.018, 0.3, 0.026}. In addition,
Table 10
The cost and technical ability for each PC.
PCk C̃2,k -cut 2,k
PC1 (0.2, 0.3, 0.4) [0.2 + 0.1, 0.4 − 0.1] 1

PC2 (0.2, 0.3, 0.4) [0.2 + 0.1, 0.4 − 0.1] 0.95
PC3 (0.4, 0.5, 0.6) [0.4 + 0.1, 0.6 − 0.1] 1
PC4 (0.3, 0.4, 0.5) [0.3 + 0.1, 0.5 − 0.1] 0.9
PC5 (0.3, 0.4, 0.5) [0.3 + 0.1, 0.5 − 0.1] 0.95
Table 11
The outcomes of the satisfaction degree and fulfillment level of each PC.
Z 2L Z 2U L
x2,1 U
x2,1 L
x2,2 U
x2,2 L
x2,3 U
x2,3 L
x2,4 U
x2,4 L
x2,5 U
x2,5
1 0.719 0.719 1 1 0.95 0.95 0.650 0.650 0.9 0.9 0.95 0.95
0.95 0.688 0.755 1 1 0.95 0.95 0.617 0.681 0.9 0.9 0.95 0.95
9 0.655 0.789 1 1 0.95 0.95 0.584 0.712 0.9 0.9 0.95 0.95
0.85 0.623 0.824 1 1 0.95 0.95 0.584 0.741 0.873 0.9 0.95 0.95
suppose that the QFD team is not concerned with the preemptive priority of PCs in this phase. In solving Model (15),
the settings of levels less than 0.85 will produce infeasible regions. This means that the different confidence levels
should be adopted for both phases in this QFD design. Applying three levels (0.85, 0.9, 0.95, 1) to Model (15), the
total DR satisfaction degree and the fulfillment level of each PC are shown in Table 11. To be in accordance with
Kano’s categorization, the corresponding business competition is changed as 2,1 = 2,1 = 1, 2,2 = 2,2 = 0.95,
2,3 = 2,3 = 1, 2,4 = 0.81, and 2,5 = 2,5 = 0.95, in constructing the fuzzy nonlinear model. In this example, PC1
should have the better quality level in order to achieve the CRs’ satisfaction degree in phase 2 of the QFD processes.
On the other hand, PC3 is smaller than the others, due to the high incremental cost, as shown in Table 10.
As described previously, the satisfaction of DR j in phase 2 should be greater than or equal to the contribution of DR j
in phase 1 in order to realize customer satisfaction. For simplification, considering the fifth DR inthe example, = 1.0,
qk
)L · x2,k qk
the satisfaction degree is equal to 0.062 by calculating either k2,5 [ k m(R2,5k ] or k2,5 [ k m(R2,5k · x 2,k ],
)U
)L = m(R2,5k q5
since m(R2,5k
at = 1.0. The fulfillment level ascribed to the PCs’ contribution is then x̂ 1,5 =
)U

k2,5 [ k m(R2,5k
q
)L · x2,k
k
]/W1,5 = 0.424 where q5 = 2 due to Kano’s “Must-be” category and W1,5 = 0.091.
q5 q5
Obviously, x̂1,5 is greater than x1,5 (0.42 = 0.16) determined in phase 1, if referring to Table 6. The resulting fulfillment
q5
level is acceptable because the possible range of x 1,5 is in [0.16, 1] (see Table 5).
5.3. Discussion
Both fuzzy nonlinear models (Models (8) and (15)) are solved by Lingo 9.0 software in a Microsoft Windows XP
environment running on a laptop with a 1.5 GHz processor as well as 512 MB random access memory. The efficiency
is satisfactory, since the software runtime is almost negligible for each -cut. In this example, the settings of levels
less than 0.4 and 0.85 will produce infeasible regions in solving fuzzy nonlinear models in phases 1 and 2 of the QFD
processes, respectively. Furthermore, the satisfaction of “what” in the latter phase should be greater than that of “how”
in the former phase to realize the most important value (customer satisfaction) that could be achieved in the QFD
processes during the design stage.
This paper considers Kano’s idea into the phases 1 and 2 models to reinforce the contribution of critical “how”
factors in the two phases. For comparisons, fuzzy linear models can be formulated by removing Kano’s idea from
fuzzy nonlinear models (Models (8) and (15)). Following the same solution procedures, fuzzy linear models for
phases 1 and 2 can be resolved using the identical parameter settings. For two kinds of models, Fig. 8 indicates
the membership functions of the satisfaction degree of phase 1 and phase 2 of the QFD processes as levels in
~ ~
~ Z1NL ~ Z2NL
1 Z1L Z2L
Member ship degree

0.95
0.9
0.85
0.3 0.4 0.5 0.6 0.7 0.8
Fig. 8. The membership functions of fuzzy nonlinear and linear models.
the feasible regions. Obviously, the satisfaction degrees illustrated in the figure from two kinds of models meet the
requirement in new product design. However, the membership functions of the satisfaction degree from fuzzy nonlinear
models are better than those from fuzzy linear models, although all of them meet the requirement in the new product
design.
As mentioned before, usually the information is insufficient at the new product design stage, such that fuzzy ap-
proaches are suitable for constructing the mathematical models. In the model-building processes, some fuzzy evalua-
tions, such as the fuzzy importance score k̃u,i and the maximum fuzzy risk priority number (RPN ) j , are defuzzified
for simplifying the models, although some information will be lost. This is because taking all fuzzy evaluations into
consideration will make the models significantly complicated. In addition, we may be interested in the case that only
crisp numbers are adopted for the evaluations in the proposed models. Actually, crisp numbers can be treated as de-
generated fuzzy numbers with equivalent lower and upper bounds of cut, i.e., level = 1. Regarding the example
demonstrated in this paper, the outcomes under such a condition can be found from the associated tables as level
equal to one. Obviously, fuzzy numbers can provide the designers with more information.
6. Concluding remarks
Determining the fulfillment levels of the “how” is an important task in the first two phases of the QFD processes,
when QFD is applied to the new product design. However, most related studies have only focused on the priority or
achievement levels of DRs in phase 1. Unlike the existing research works, both phases 1 and 2 are considered in this
paper. Considering Kano’s concept, this paper proposes a fuzzy nonlinear model to determine the performance levels
of the “how” based on the existing fuzzy linear model. The “how” can be categorized into three classes based on Kano’s
concept. Through the modification of the decision variables denoting the performance level, the fuzzy nonlinear models
can be constructed for phases 1 and 2 in a fuzzy environment. Compared with fuzzy linear models without considering
Kano’s concept, fuzzy nonlinear models can achieve better satisfaction degree.
To reduce the design risk, risk analysis of DRs, namely a fuzzy FMEA, is taken into account in the phase 2 model as
a constraint factor. In addition, this paper considers the close link between the two phases to build up a fuzzy nonlinear
programming model for phase 2 in determining the fulfillment levels of PCs. The satisfaction of DR j determined in
phase 2 is greater than or equal to the contribution of DR j in phase 1 in order to realize customer satisfaction. The
proposed model is illustrated with a numerical example to demonstrate its applicability in practice. The resulting ranges
of satisfaction degrees and the possible ranges of fulfillment levels of the “how” in both phases can provide the QFD
team with useful information for new product design. Besides Kano’s concept and FMEA, in future research other
considerations, such as means-end chain (MEC), the integration of more phases of QFD activity, and so on, could be
modeled in the QFD processes for the new product design and development. In addition, more elaborate approaches
could be developed to solve the problem, in which all fuzzy evaluations are used as inputs in the models in order to
obtain complete information.
Acknowledgements
The authors appreciate the anonymous reviewer’s useful comments to make this paper more comprehensive. This
research was financially supported by the National Science Council, Republic of China, under contract NSC95-2416-
H-006-028-MY3.
Appendix A
The fuzzy Delphi method consists of four steps to determine the importance score of CRs, and is described as follows
[18,19].
Step 1: The current and potential customers, acting as an expert role Custm , m = 1, 2, . . . , n, provide the possible
importance score of CRi , i = 1, 2, . . . , I . The importance score assigned by each expert Custm is presented in the form
of a triangular fuzzy number as
k̃i,m = [(ki,m )a , (ki,m )b , (ki,m )c ], m = 1, 2, . . . , n; i = 1, 2, . . . , I, (A.1)
where k̃i,m denotes the importance score of CRi by the expert m, the indexes “a”, “b”, “c” represent the lowest, the
most likely and the highest level of k̃i,m , respectively.
Step 2: Computing the average of all k̃i,m , the formulation is expressed as

1
n n n
a 1 b 1
k̄i = [(k̄i ) , (k̄i ) , (k̄i ) ] =
a b c
(ki,m ) , (ki,m ) , (ki,m ) ,
c
(A.2)
n n n
m=1 m=1 m=1
where i = 1, 2, . . . , I . Then, for each expert Custm the differences between k̃i,m and k̄ i are calculated and returned to
each expert for the next re-evaluation run.
Step 3: Each expert takes the differences in the previous step as a reference, and then provides a revised linguistic
variable (if necessary) of the importance scores as

k̃i,m = [(ki,m )a , (ki,m )b , (ki,m )c ], m = 1, 2, . . . , n; i = 1, 2, . . . , I. (A.3)
Then the average is calculated using a distance measure,

can also be obtained from (A.2) and its difference with k̄
k̄i,m i
proposed by Bojadziev and Bojadziev [3], as follows:
d(k̄ i , k̄ i ) = 0.5{max[|(k̄ i )a − (k̄ i )a |, |(k̄ i )c − (k̄ i )c |] + |(k̄ i )b − (k̄ i )b |}. (A.4)

This process could be repeated until two adjacent averages become reasonably close. The consensus criteria is assumed
to be a small distance, such as d 0.2.
Step 4: If there is any change or updated information that may lead to a re-evaluation of the importance scores of
CRs at a later time, Step 1 should start the next-round process.
Appendix B
The fuzzy Delphi method is applied to determine the importance score of CRs in T2 -BGA package development.
The computational procedure is described as follows:
Step 1: It is assumed that a QFD team forwards a questionnaire with linguistic evaluation levels, “very low (VL)”,
“low (L)”, “medium (M)”, “high (H)” or “very high (VH)”, to eight current and potential customers in order to obtain
the importance scores of five CRs, as mentioned in Section 4. As the experts, the eight customers use those five linguistic
terms to express the importance score of each CR. Their first evaluations are listed in Table B1. The QFD team then
translates those linguistic terms into the triangular fuzzy numbers as a 3-element set, namely (0, 0, 0.2), (0, 0.2, 0.4),
(0.3, 0.5, 0.7), (0.6, 0.8, 1), and (0.8, 1, 1), respectively. For example, the importance score of the fifth CR evaluated
by the second expert is expressed as k̃5,2 = (0, 0.2, 0.4).
Step 2: Computing the average of all k̃i,m by (A.2), the results of k̄ i produced for each CR are (0.075, 0.225, 0.425),
(0.263, 0.463, 0.663), (0.113, 0.263, 0.463), (0.438, 0.638, 0.813), and (0.038, 0.138, 0.338), respectively. Then, the
differences between k̃i,m and k̄ i are calculated and sent back to each corresponding expert for re-evaluation.
Table B1
The first round evaluation of importance score of the each CR.
Expert CR1 CR2 CR3 CR4 CR5
Cust 1 L M L H VL
Cust 2 L M VL H L
Cust 3 VL H M M L
Cust 4 M L L H M
Cust 5 L L L VH VL
Cust 6 L M H M VL
Cust 7 M M L L VL
Cust 8 VL M VL M L
Table B2
The second round evaluation of importance score of the each CR.
Expert CR1 CR2 CR3 CR4 CR5
Cust 1 L M L H L
Cust 2 L M L H L
Cust 3 L M M M L
Cust 4 L L L H L
Cust 5 L M L VH L
Cust 6 L M H H VL
Cust 7 M M L M VL
Cust 8 L M L H L
can also be
Step 3: As in Step 1, each expert provides a revised evaluation, as listed in Table B2. The average of k̃i,m
obtained from (A.2), and then the k̄ of each CR is determined as (0.038, 0.238, 0.438), (0.263, 0.463, 0.663), (0.113,
i
0.313, 0.513), (0.55, 0.75, 0.925), and (0, 0.15, 0.35), respectively. The distance between k̄ i and k̄ i is calculated by (A.4)
as 0.025, 0, 0.05, 0.113, 0.025, respectively. Since the distances between two adjacent averages are less than or equal
to 0.2, this means that the consensus condition is achieved. The resulting fuzzy number k̄ i will be applied to determine
the fuzzy technical importance ratings. If the consensus condition was not obtained, the re-evaluation process would
go on to the next round until the consensus condition is achieved.
References
[1] Y. Akao, G.H. Mazur, The leading edge in QFD: past, present and future, International Journal of Quality & Reliability Management 20 (2003)
20–35.
[2] M. Al-Mashari, M. Zairi, D. Ginn, Key enablers for the effective implementation of QFD: a critical analysis, Industrial Management & Data
Systems 105 (2005) 1245–1260.
[3] G. Bojadziev, M. Bojadziev, Fuzzy Sets, Fuzzy Logic, Applications—Advances in Fuzzy Systems—Applications and Theory, Vol. 5, World
Scientific, Singapore, 1995.
[4] J. Bondia, J. Picó, Analysis of linear systems with fuzzy parametric uncertainty, Fuzzy Sets and Systems 135 (2003) 81–121.
[5] L.K. Chan, M.L. Wu, A systematic approach to quality function deployment with a full illustrative example, Omega—The International Journal
of Management Science 33 (2005) 119–139.
[6] Y. Chen, R.Y.K. Fung, J. Tang, Rating technical attributes in fuzzy QFD by integrating fuzzy weight average method and fuzzy expected value
operator, European Journal of Operational Research 174 (2006) 1553–1566.
[7] K.M. Chen, K.H. Horng, K.N. Chiang, Coplanarity analysis and validation of PBGA and T2 -BGA packages, Finite Elements in Analysis and
Design 38 (2002) 1165–1178.
[8] L.H. Chen, W.C. Ko, A fuzzy nonlinear model for quality function deployment considering Kano’s concept, Mathematical and Computer
Modeling 48 (2008) 581–593.
[9] L.H. Chen, M.C. Weng, A fuzzy model for exploiting quality function deployment, Mathematical and Computer Modeling 38 (2003) 559–570.
[10] L.H. Chen, M.C. Weng, An evaluation approach to engineering design in QFD processes using fuzzy goal programming models, European
Journal of Operational Research 172 (2006) 230–248.
[11] L. Cohen, Quality Function Deployment: How to Make QFD Work for You, Addison Wesley, Reading, MA, 1995.
[12] J.J. Cristiano, J.K. Liker, C.C. White, III, Key factors in the successful application of quality function deployment (QFD), IEEE Transactions
on Engineering Management 48 (2001) 81–95.
[13] D.M. Ginn, D.V. Jones, H. Rahnejat, M. Zairi, The QFD/FMEA interface, European Journal of Innovation Management 1 (1998) 7–20.
[14] A.C.F. Guimarães, C.M.F. Lapa, Fuzzy FMEA applied to PWR chemical and volume control system, Progress in Nuclear Energy 44 (2004)
191–213.
[15] S.B. Han, S.K. Chen, M. Ebrahimpour, M.S. Sodhi, A conceptual QFD planning model, The International Journal of Quality & Reliability
Management 18 (2001) 796–812.
[16] J.R. Hauser, D. Clausing, The house of quality, Harvard Business Review 66 (1988) 63–73.
[17] N. Kano, N. Seraku, F. Takahaashi, S. Tsuji, Attractive quality and must-be quality, Hinshitsu: The Journal of the Japanese Society for Quality
Control (1984) 39–48.
[18] E.E. Karsak, Fuzzy multiple objective decision making approach to prioritize design requirements in quality function deployment, International
Journal of Production Research 42 (2004) 3957–3974.
[19] A. Kaufmann, M.M. Gupta, Fuzzy Mathematical Models in Engineering and Management Science, Elsevier Science Inc., New York, NY, 1988.
[20] G. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logic: Theory and Application, Prentice-Hall, NJ, 2003.
[21] C.K. Kwong, H. Bai, Determining the importance weights for the customer requirements in QFD using a fuzzy AHP with an extent analysis
approach, IIE Transactions 35 (2003) 619–626.
[22] T. Lager, The industrial usability of quality function deployment: a literature review and synthesis on a meta-level, R&D Management 35 (2005)
409–426.
[23] W.V. Leekwijck, E.E. Kerre, Defuzzification: criteria and classification, Fuzzy Sets and Systems 108 (1999) 159–178.
[24] S. Mabuchi, A proposal for a defuzzification strategy by the concept of sensitivity analysis, Fuzzy Sets and Systems 55 (1993) 1–14.
[25] K. Matzler, H.H. Hinterhuber, How to make product development projects more successful by integrating Kano’s model of customer satisfaction
into quality function deployment, Technovation 18 (1998) 25–38.
[26] S. Myint, A framework of an intelligent quality function deployment (IQFD) for discrete assembly environment, Computers & Industrial
Engineering 45 (2003) 269–283.
[27] A. Pillay, J. Wang, Modified failure mode and effects analysis using approximate reasoning, Reliability Engineering & System Safety 79 (2003)
69–85.
[28] R.K. Sharma, D. Kumar, P. Kumar, Systematic failure mode effect analysis (FMEA) using fuzzy linguistic modeling, The International Journal
of Quality & Reliability Management 22 (2005) 986–1004.
[29] D.H. Stamatis, Failure Mode and Effect Analysis—FMEA from Theory to Execution, ASQC Press, New York, 1995.
[30] C.M. Tan, Customer-focused build-in reliability: a case study, The International Journal of Quality & Reliability Management 20 (2003)
378–397.
[31] K.C. Tan, T.A. Pawitra, Integrating SERVQUAL and Kano’s model into QFD for service excellence development, Managing Service Quality
11 (2001) 418–430.
[32] K.C. Tan, X.X. Shen, Integrating Kano’s model in the planning matrix of quality function deployment, Total Quality Management 11 (2000)
1141–1151.
[33] G.S. Wasserman, On how to prioritize design requirements during the QFD planning process, IIE Transactions 25 (1993) 59–65.
[34] H.C. Wu, Linear regression analysis for fuzzy input and output data using the extension principle, Computers and Mathematics with Applications
45 (2003) 1849–1859.
[35] Z.S. Xu, Q.L. Da, An overview of operators for aggregating information, International Journal of Intelligent Systems 18 (2003) 953–969.
[36] K. Xu, L.C. Tang, M. Xie, S.L. Ho, M.L. Zhu, Fuzzy assessment of FMEA for engine systems, Reliability Engineering & System Safety 75
(2002) 17–29.
[37] L.A. Zadeh, Fuzzy set as a basis for a theory of possibility, Fuzzy Sets and Systems 1 (1978) 3–28.
[38] H.J. Zimmermann, Fuzzy Set Theory and its Applications, second ed., Kluwer-Nijhoffg, Boston, MA, 1991.

Fuzzy Approaches To Quality Function Deployment For New Product Design

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fuzzy Approaches To Quality Function Deployment For New Product Design

Uploaded by

Copyright:

Available Formats

Fuzzy Sets and Systems 160 (2009) 2620 – 2639

Fuzzy approaches to quality function deployment for new

∗ Corresponding author. Fax: +886 6 236 2162.

DR1 . . . DRj . . . DRJ PC1 . . . PCk . . . PCK

CR1 k1,1 DR1 k2,1

(R1,i )L = inf {x| R̃1,i (x) } and (2a)

= sup {x| R̃1,i (x) },

function at each -cut are formulated as

3. Fuzzy formulations of phase 1

Fig. 2. Kano’s conceptual model.

where B1 is the budget limitation, and (Z 1 )L as well as (Z 1 )U

4. Fuzzy formulations of phase 2

Frequency of occurrence Rating Possible failure rate

Remote 1 < 1 : 20, 000

Detectability Rating Probability of detectability (%)

where (RPNj )q = 21 [(RPNj )Lq + (RPNj )U

Chip Goldwire Heatslug Molding compound

Solder ball Substrate

Fig. 3. The cross-sections of PBGA and T2 -BGA.

DR1 DR2 DR3 DR4 DR5

Fig. 4. The phase 1 HOQ for a T2 BGA package.

Fig. 5. The membership functions of linguistic variables in the HOQ.

5.1. Phase 1 of the QFD process

DR j C̃1, j -cut 1, j 1, j

1 0.559 0.559 1 1 0.151 0.151 0.515 0.515 1 1 0.4 0.4

5.2. Phase 2 of the QFD process

Material attached to Molding failure Package profile defect VH Turbulence M Simulation H

Molding flow Popcorn Reliability test failed H Turbulence M Simulation L

Remote Low Moderate High Veryhigh

Fig. 6. The membership functions of linguistic terms in the FMEA.

PC1 PC2 PC3 PC4 PC5

Fig. 7. The phase 2 HOQ for a T2 BGA package.

PCk C̃2,k -cut 2,k

PC1 (0.2, 0.3, 0.4) [0.2 + 0.1, 0.4 − 0.1] 1

Member ship degree

Fig. 8. The membership functions of fuzzy nonlinear and linear models.

Then the average is calculated using a distance measure,

d(k̄ i , k̄ i ) = 0.5{max[|(k̄ i )a − (k̄ i )a |, |(k̄ i )c − (k̄ i )c |] + |(k̄ i )b − (k̄ i )b |}. (A.4)

Expert CR1 CR2 CR3 CR4 CR5

Expert CR1 CR2 CR3 CR4 CR5

You might also like

DR j C̃1, j -cut 1, j 1, j

PCk C̃2,k -cut 2,k